Interactive Reconstruction of 3-D Objects from Silhouettes

Abstract

Many techniques for reconstructing 3-D shapes from 2-D images use silhouette data. A problem with this approach is that, if no a priori information about the 3-D shape is available, we do not know neither the accuracy of the reconstruction, nor where it is better to locate new viewpoints for improving the accuracy. We present and demonstrate a new general approach to interactive, object-specific shape-from-silhouette algorithms. The approach holds for completely unknown shapes. It is based on a necessary condition for the reconstruction to have been performed with the best possible accuracy. From this condition, we derive: 1) a quantitative measure of reconstruction accuracy; 2) rules for finding new viewpoints if the accuracy is not satisfactory. The algorithm has been implemented for polyhedra, and demonstrated in a virtual environment.